資源描述:
《基于改進(jìn)型模糊神經(jīng)網(wǎng)絡(luò)的智能控制研究》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、摘要模糊神經(jīng)網(wǎng)絡(luò)著眼于模糊控制的邏輯推理技術(shù)和神經(jīng)網(wǎng)絡(luò)控制的自學(xué)習(xí)能力,試圖將二者巧妙地結(jié)合,是智能控制在近年來涌現(xiàn)出的重要研究領(lǐng)域之一,具有學(xué)術(shù)理論意義和實際應(yīng)用價值。目前,已有多種形式的模糊神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu),但主要表現(xiàn)為將兩者在結(jié)構(gòu)和算法上相互融合。本文在深入研究模糊神經(jīng)網(wǎng)絡(luò)發(fā)展的基礎(chǔ)上,提出了將模糊控制中的部分控制理念和方法應(yīng)用到神經(jīng)網(wǎng)絡(luò)中的構(gòu)想,并通過仿真驗證了可行性,為模糊神經(jīng)網(wǎng)絡(luò)的發(fā)展開辟了新的思路。首先,在對當(dāng)前模糊神經(jīng)網(wǎng)絡(luò)發(fā)展領(lǐng)域進(jìn)行深入研究后,提出了自己的新型分類研究方法,即從神經(jīng)網(wǎng)絡(luò)和模糊邏輯推理的分類角度剖析模糊神經(jīng)網(wǎng)絡(luò)的結(jié)構(gòu),并在算法研
2、究中介紹了一種新的可行性尋優(yōu)算法——微粒群算法。其次,詳細(xì)介紹了一種典型的新型網(wǎng)絡(luò)——正則化模糊神經(jīng)網(wǎng)絡(luò),并進(jìn)行了具體的算法推導(dǎo),還利用StoneWeierstrass定理對網(wǎng)絡(luò)收斂性進(jìn)行了清晰地證明,并引用具體的辨識問題對其性能加以驗證。再次,給出了一種基于RBF辨識的改進(jìn)型模糊神經(jīng)網(wǎng)絡(luò)控制,即將量化因子和比例因子嵌入到模糊神經(jīng)網(wǎng)絡(luò)中,構(gòu)成網(wǎng)絡(luò)的輸入層和輸出層;并采用RBF辨識網(wǎng)絡(luò)取代傳統(tǒng)的近似做法,為系統(tǒng)提供精確的Jacobian信息;最后,在仿真過程中,還引出了一種對基本論域和模糊論域有效轉(zhuǎn)化的理想方案。最后,受模糊控制中“調(diào)整因子”的啟發(fā),給出了一
3、種含有調(diào)整因子的模糊RBF網(wǎng)絡(luò),并采用微粒群優(yōu)化算法對調(diào)整因子進(jìn)行滾動優(yōu)化調(diào)整。關(guān)鍵詞微粒群算法;優(yōu)化;神經(jīng)網(wǎng)絡(luò);模糊系統(tǒng);模糊神經(jīng)網(wǎng)絡(luò)燕山大學(xué)工學(xué)碩士學(xué)位論文AbstractFuzzyneuralnetworkfocllsesbothonlogicreasoningtechnologyoffuzzycontrolandself-learningcapabilityofneuralnetworkcontrol,andtriestomakeappropriateintegrationofthoseadvantages,whichisoneoftheeme昭e
4、ntlyimportantresearchesintheintelligentcontrolnearlyseveralyears.Alltheseresearchesareofprofoundtheorysignificanceandpracticalapplicationvalue.Therearemanykindsoffuzzyneuralnetworkspresently.But,theymanifestmainlyonthecombinationofframeandarithmetic.Basedonthestudyoffuzzyneuralnet
5、work’Sdevclopment,thispaperproposedtheideathatappliedsometheoriesandmethodsoffuzzycontroltoneuralnetwork,andconfirmedthefeasibilitybysimulation,whichopenednewwaysforfuzzyneuralnetwork’sdevelopment.Firstly,afterfurtherresearchonthetemporaryfuzzyneuralnetwork’Sdevelopment,anewclassi
6、ficationmethodwasproposed.Itanalyzedthefuzzyneuralnetwork’Sframefromthepointofclassificationofneuralnetworkandfuzzylogicreasoning.And,thereintroducedanewfeasibleoptimizationarithmeticinthearithmeticresearch.whichwascalledparticleswamioptimizationarithmetic.Secondly,atypicallynewne
7、tworkwhichWascallednormalfuzzyneuralnetworkwasproposedindetail.Itgaveconcretearithmeticandcleardemonstrationaboutthenetwork’Sconvergencebytheorem—StoneWeierstrass.Anditsperformancewasconfn-medbyconcreteidentification.Thirdly,thereintroducedallimprovedfuzzyneuralnetworkcontrolbased
8、onidentificationofRBRThescalingfa